NASA and IBM have teamed up on an exciting new collaboration – developing an open-source artificial intelligence (AI) model to derive key insights from NASA satellite data about our planet. This NASA and IBM AI partnership combine NASA’s wealth of Earth observation data with IBM’s AI capabilities on the IBM Cloud and supercomputing infrastructure.
The joint project aligns with NASA’s commitment to open science and reflects the vast potential of AI to accelerate climate and geospatial research. By training the model on NASA’s Harmonised Landsat Sentinel-2 (HLS) dataset, and then openly sharing it, the partners aim to empower the scientific community to tap into sophisticated AI to better understand Earth processes.
An Overview of the NASA-IBM Partnership
NASA and IBM formally announced their collaboration in November 2022. The foundation of the partnership is an AI model trained on NASA’s HLS dataset on IBM’s Vela supercomputer.
The HLS dataset combines observations from multiple Earth observation satellites. This provides a rich corpus for the AI model to learn from.
The model development leverages IBM’s Harrisson AI platform, including workflows and experiment management tools. It runs on the IBM Cloud within a scalable, secure environment.
By open-sourcing the model, NASA and IBM hope to make such advanced AI accessible to more researchers. This aligns with NASA’s 10-year Open Science plan to build an open, collaborative science community.
Capabilities of the Joint Geospatial AI Model
The AI model developed by NASA and IBM scientists has impressive capabilities when applied to the analysis of geospatial data:
- Interprets satellite images up to 4 times faster than existing models, enabling quick insights.
- Requires less labeled training data to reach high accuracy – reducing time and costs.
- Can track changes over time in land use, vegetation, infrastructure, disaster damage, and more based on satellite imagery.
- Helps predict outcomes like crop yields, flood risks, and deforestation by revealing trends in image data.
- Creates visualizations from the images to spot patterns and surface anomalies for further investigation.
Later in 2023, IBM’s Environmental Intelligence Suite will offer a commercial edition of the model, ensuring widespread availability.
Also Read: Is NightCafe AI-Created Artwork Really Legit? Exploring the Legitimacy of AI-Generated Creativity
The HLS Dataset Fuels AI’s Geospatial Learning
The Harmonised Landsat Sentinel-2 (HLS) dataset compiled by NASA and the ESA underpins the model’s training. This dataset combines readings from multiple Earth observation satellites:
- Landsat 8 and Landsat 9, jointly managed by NASA and the USGS
- Sentinel-2 satellites, part of the European Union’s Copernicus Earth observation program
The coordinated observations from these satellites enable consistent surface imaging at a 30-meter resolution, with global coverage every 2-3 days.
This provides a rich, updated corpus of Earth imagery for the AI algorithm to learn from. When ingested into the model, it can start recognizing patterns and changes on the Earth’s surface.
Advancing the Model Through Collaboration
To refine the AI model, NASA and IBM are facilitating workshops with partner organizations that can provide domain expertise and evaluate real-world performance.
Partners engaged so far include the IEEE Geoscience and Remote Sensing Society, the Earth Science Informatics Technical Committee, and Oak Ridge National Laboratory.
Feedback and testing from different groups allow the model to be fine-tuned for optimal effectiveness across diverse use cases. The collaborators can also suggest helpful new training data to integrate.
This collaborative approach follows NASA’s principles of open science. Ongoing improvement will help the model reach its full potential as a tool for accelerating discovery.
Also Read: Is NightCafe AI-Created Artwork Really Legit? Exploring the Legitimacy of AI-Generated Creativity
The Road Ahead for AI-Enhanced Earth Science
The NASA and IBM AI partnership powerfully demonstrates artificial intelligence’s potential to uncover new insights from Earth observation data. By openly sharing a sophisticated model trained on rich NASA data, they break down barriers for researchers to tap into leading-edge AI capabilities.
As the model continues to evolve through collaboration, it could help unlock scientific advances in climate modeling, agriculture, disaster response, and environmental monitoring. More broadly, it highlights the growing role of AI in enabling evidence-based decisions about how to protect our planet.
